https://doi.org/10.5061/dryad.b5mkkwhnw

Description of the Data and File Structure

Trust Region Bayesian Optimized Network of Artificial Olfactory Receptors for Spatiotemporal Monitoring of NO₂ Gas

This repository contains supplementary codes and datasets for the paper:

Trust Region Bayesian Optimized Network of Artificial Olfactory Receptors for Spatiotemporal Monitoring of NO₂ Gas.

The code includes TFLite-based file generation, dataset handling, testing, and validation processes.

Note: Before starting, ensure you have Jupyter Notebook or Jupyter Lab installed for running the .ipynb files.

Dataset

Find the attached dataset files.

Files and Variables

File: Datasets.npy

Description: Contains gas simulation datasets for the TuRBO optimization and neural network training.

The dimension of dataset is (n × 11, 4455, 3). The dataset represent the contour map of gas concentration throughout the observation space, which is 50 × 23 (m) simulative space. The ‘n’ number of observations were taken which represent each gas leakage scenarios with different gas leakage points, observed for 11 times within 1000 seconds (0, 100, 200, … , 1000). The second shape parameter (4455) represents the all points in the contour map. The last dimension 3 represents x, y, and concentration. Specifically, the first 2 of 3 contains x and y coordinate of the observation point, and the last 1 contains the concentration level.

• Shape: (n × 11, 4455, 3)
  • n × 11: Total number of observations multiplied by 11 time steps (from 0 to 1000 seconds at 100-second intervals).
  • 4455: Number of points in the contour map.
  • 3: Represents [x, y, concentration] at each point.

Visualization Example:

To visualize the concentration map for the 15th observation at the 500-second timestamp, use the following code:

import numpy as np
import matplotlib.pyplot as plt

# Load the dataset
DATASET = np.load('Datasets.npy')

# Parameters
observation_number = 15  # 15th observation
time_step = 500          # Time in seconds

# Time steps in the dataset
time_steps = [0, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000]
time_index = time_steps.index(time_step)  # Get the index of the desired time step

# Calculate the dataset index
index = (observation_number - 1) * 11 + time_index

# Extract x, y coordinates and concentration values
x_coords = DATASET[index, :, 0]
y_coords = DATASET[index, :, 1]
concentration = DATASET[index, :, 2]

# Plotting
plt.figure(figsize=(10, 8))
scatter = plt.scatter(x_coords, y_coords, c=concentration, cmap='viridis')
plt.colorbar(scatter, label='Concentration Level')
plt.xlabel('X Coordinate (m)')
plt.ylabel('Y Coordinate (m)')
plt.title(f'Concentration Map for Observation {observation_number} at {time_step} sec')
plt.show()

File: Datalabels.npy

Description: Contains the labels for each dataset entry. This file is embedded in the code folder found on Zenodo.

• Shape: (n × 11, 2)
  • The last dimension [2] represents the [x, y] coordinates of the gas leakage point.

Code/Software

Basic Jupyter Notebook Setup and Run Guide

Prerequisites

Ensure you have the following installed on your system:

• Python 3.12
• pip (Python package installer)

Installation

1. Install Jupyter Notebook via pip

   Open your terminal (for macOS/Linux) or Command Prompt/PowerShell (for Windows) and run the following command:

   pip install notebook
   

2. Verify Installation

   To ensure Jupyter Notebook was installed correctly, run:

   jupyter --version
   

   This should display the version of Jupyter Notebook installed.

Running Jupyter Notebook

1. Start Jupyter Notebook

   Navigate to the directory where your notebooks are located or where you wish to create new ones, then run:

   jupyter notebook
   

   This command will start the Jupyter Notebook server and open the interface in your default web browser.

2. Create a New Notebook
   • In the Jupyter Notebook interface, click the “New” button at the top right and select “Python 3” to create a new notebook.
3. Using Jupyter Notebook
   • Running Cells: Write Python code in the cells and run them interactively by pressing Shift + Enter.
   • Managing Cells: Add new cells, delete cells, or change the cell type (Code or Markdown) using the toolbar.
4. Save and Exit
   • Save: Click the floppy disk icon in the toolbar to save your notebook.
   • Exit: Close the browser tab and stop the Jupyter server in your terminal by pressing Ctrl + C.

Additional Resources

For more detailed instructions and advanced usage, visit the Jupyter Project Documentation.

Run the Code

1. Run BO_Gas_sensor_T9_EI-TurBO_10.ipynb

This notebook performs Bayesian Optimization to generate an optimized neural network and positions 10 sensors in optimized regions for better gas detection.

Steps:

• Open the notebook in Jupyter.
• Run each cell sequentially by pressing Shift + Enter.
• The notebook includes code for data loading, preprocessing, optimization, and visualization.

2. Run SPresense_TF_GAS.ino

This code sets the Sony SPresense microprocessor into listening mode for gas concentration data.

Instructions:

• Setup: Refer to the following resources for detailed settings and environment setup:
  • TFLite Micro SPresense Examples on GitHub
  • Running TensorFlow Lite on the Sony Spresense
• Upload: Use the Arduino IDE or Sony’s development environment to upload the .ino file to the SPresense board.

3. Run BO_Gas_sensor_T9_Neural_Network_verify_Board.ipynb

This notebook trains and validates the neural network based on the sensor positions generated by the Bayesian Optimization.

Steps:

• Open the notebook in Jupyter.
• Ensure that the optimized sensor positions from the previous steps are correctly loaded.
• Run the training and validation cells to evaluate the neural network’s performance.

Access Information

Other publicly accessible locations of the data:

• https://github.com/bb8ax/TuRBO_Gas/

Data was derived from the following sources:

• None